{"slug": "orbital-optimization-of-robust-brain-computer-inference-through-architecture-low", "title": "ORBITAL: Optimization of Robust Brain--Computer Inference Through Architecture-Aware Low-Precision Compression", "summary": "A framework called ORBITAL, presented in arXiv paper 2610.36411v1, produced S25-TRADES, which achieved the highest observed mean true-union robustness among compressed EEGNet models on BCI Competition IV-2a at 23.06%, versus 21.27% for standard Full, while using 26.27% fewer parameters and 25.10% fewer MACs across nine subjects and three seeds. Matched controls identified TRADES as the main source of robustness recovery, improving S25 union accuracy by 6.74 percentage points over clean recovery (p=0.0039), and on Jetson Orin Nano FP16 compression cut board energy by up to 16.20% and latency by 8.00%. Despite 99.3% prediction agreement, native INT8 Q/DQ incurred about twice the FP16 latency and energy, which the authors cite as evidence that robustness and hardware-aware optimization must be tuned jointly for embedded brain-computer interface decoders.", "body_md": "arXiv:2610.36411v1 Announce Type: new \nAbstract: Compact electroencephalography (EEG) decoders enable embedded brain--computer interfaces (BCIs), but compression can alter adversarial robustness without proportional hardware savings. We present ORBITAL, a robustness-constrained framework combining hardware-realizable structured compression, recovery-aware operating-point selection, multi-attack evaluation, and native edge validation. Across nine subjects and three seeds on BCI Competition IV-2a, S25-TRADES achieves the highest observed mean true-union robustness among compressed EEGNet models (23.06% vs 21.27% for standard Full), with 26.27% fewer parameters and 25.10% fewer MACs. Matched controls identify TRADES as the main source of robustness recovery, improving S25 union accuracy by 6.74 percentage points over clean recovery (p=0.0039), while per-example analysis reveals boundary redistribution rather than uniform margin expansion. On Jetson Orin Nano, FP16 compression reduces board energy by up to 16.20% and latency by 8.00%. Despite 99.3% prediction agreement, native INT8 Q/DQ incurs about twice the FP16 latency and energy, demonstrating the need for joint robustness and hardware-aware optimization.", "url": "https://wpnews.pro/news/orbital-optimization-of-robust-brain-computer-inference-through-architecture-low", "canonical_source": "https://arxiv.org/abs/2610.36411", "published_at": "2026-10-28 04:00:00+00:00", "updated_at": "2026-10-28 04:14:41.012166+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "ai-infrastructure", "neural-networks"], "entities": ["ORBITAL", "EEGNet", "S25-TRADES", "BCI Competition IV-2a", "Jetson Orin Nano", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/orbital-optimization-of-robust-brain-computer-inference-through-architecture-low", "markdown": "https://wpnews.pro/news/orbital-optimization-of-robust-brain-computer-inference-through-architecture-low.md", "text": "https://wpnews.pro/news/orbital-optimization-of-robust-brain-computer-inference-through-architecture-low.txt", "jsonld": "https://wpnews.pro/news/orbital-optimization-of-robust-brain-computer-inference-through-architecture-low.jsonld"}}